A multiple range test for analysis of covariance

Abstract

A new test statistic is proposed for making multiple comparisons on adjusted treatment means in analysis of covariance when the dependent and concomitant variates have a multivariate normal distribution. The probability distribution of the range of the adjusted treatment means, suitably Studentized, is derived for the case of one-way analysis of covariance with equal subgroup sizes and one random covariate, and the distribution is found to be free of nuisance parameters. Some percentage points of the statistic are computed and presented, together with an example